MEMORY INDUSTRY INTELLIGENCE
근거
1675건 공개자료 기반 개인 관찰 기록
| 기록 | 상태·범위 | 수정일·발행일 | |
|---|---|---|---|
| For media inquiries, email [[email protected]](https://www.federalreserve.gov/cdn-cgi/l/email-protec | 초안 | 2026.10.10 20:15 | |
| For release at 2:00 p.m. EDT | 초안 | 2026.10.10 20:15 | |
| The attached tables and charts released on Wednesday summarize the economic projections made by Fede | 초안 | 2026.10.10 20:15 | |
| All runs cleared closed-division accuracy: gpt-oss-120b at 83.69% exact-match (Offline) and 83.64% ( | 초안 | 2026.10.10 20:10 | |
| By rebalancing the prefill:decode GPU split in a disaggregated NVIDIA Dynamo deployment from 75:25 t | 초안 | 2026.10.10 20:10 | |
| Scaling from one node to four was effectively linear. Qwen3-VL Server throughput at 16 GPUs came in | 초안 | 2026.10.10 20:10 | |
| Our gpt-oss-120b submission is likewise the highest per-GPU GB200 NVL72 result in the closed divisio | 초안 | 2026.10.10 20:10 | |
| NVIDIA's guidance for a 4x GB200 NVL72 node was roughly 56 samples/s Offline and 46 queries/s Server | 초안 | 2026.10.10 20:10 | |
| Today, MLCommons published the MLPerf® Inference v6.1 results, and with them Crusoe's first-ever MLP | 초안 | 2026.10.10 20:10 | |
| At Crusoe, we partner closely with AMD on hardware performance and software co-design, from ROCm and | 초안 | 2026.10.10 20:09 | |
| We will keep deepening this work with AMD and our partners, and we will keep publishing results for | 초안 | 2026.10.10 20:09 | |
| gpt-oss-120b: 8 independent single-GPU replicas per node at tensor-parallel size 1, for 512 replicas | 초안 | 2026.10.10 20:09 | |
| Both models cleared closed-division accuracy: gpt-oss-120b at 83.7% exact-match against an 82.3% ref | 초안 | 2026.10.10 20:09 | |
| The gpt-oss-120b model, with its MoE weights in MXFP4, occupies roughly 65 GB. It runs at tensor-par | 초안 | 2026.10.10 20:09 | |
| On MI355X the same model sits comfortably inside a single 8-GPU platform with 2.3 TB of HBM3E, leavi | 초안 | 2026.10.10 20:09 | |
| Throughput scales linearly from 8 to 512 GPUs at over 90% of ideal, which means workloads can be aut | 초안 | 2026.10.10 20:09 | |
| Crusoe's MLPerf Inference v6.1 submission ran gpt-oss-120b and DeepSeek-R1 on 512 AMD Instinct MI355 | 초안 | 2026.10.10 20:09 | |
| NVIDIA HGX B300 is available on Crusoe Cloud now, with trial runs before you commit and solutions en | 초안 | 2026.10.10 20:08 | |
| Crusoe's NVIDIA HGX B300 validation ran on 512 Blackwell Ultra GPUs across a broad set of frontier m | 초안 | 2026.10.10 20:08 | |
| Achieving NVIDIA Exemplar Cloud validation on geothermal-powered infrastructure, following our hydro | 초안 | 2026.10.10 20:08 | |
| The NVIDIA HGX B300 systems used for this validation ran in Iceland, powered entirely by geothermal | 초안 | 2026.10.10 20:08 | |
| NVFP4 is also more computationally efficient: 4-bit values with shared scaling drastically reduce me | 초안 | 2026.10.10 20:08 | |
| Networking bandwidth doubles to 1.6 TB/s per system, with 800 Gb/s per GPU feeding the scale-out fab | 초안 | 2026.10.10 20:08 | |
| Compared with NVIDIA HGX B200, it delivers 1.5x the dense NVFP4 compute, 2x the attention-layer perf | 초안 | 2026.10.10 20:08 | |
| Crusoe Cloud has earned NVIDIA Exemplar Cloud validation for large-scale AI training on NVIDIA HGX B | 초안 | 2026.10.10 20:08 |